The simplicity of text files is appealing - did you consider using a database at all or was it pure text from the start?

Thinking about an LLM use case, not needing a query language should remove translation risk I'd assume?